• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.1-17
• /
• 2017

The deep learning framework is software designed to help develop deep learning models. Some of its important functions include "automatic differentiation" and "utilization of GPU". The list of popular deep learning framework includes Caffe (BVLC) and Theano (University of Montreal). And recently, Microsoft's deep learning framework, Microsoft Cognitive Toolkit, was released as open-source license, following Google's Tensorflow a year earlier. The early deep learning frameworks have been developed mainly for research at universities. Beginning with the inception of Tensorflow, however, it seems that companies such as Microsoft and Facebook have started to join the competition of framework development. Given the trend, Google and other companies are expected to continue investing in the deep learning framework to bring forward the initiative in the artificial intelligence business. From this point of view, we think it is a good time to compare some of deep learning frameworks. So we compare three deep learning frameworks which can be used as a Python library. Those are Google's Tensorflow, Microsoft's CNTK, and Theano which is sort of a predecessor of the preceding two. The most common and important function of deep learning frameworks is the ability to perform automatic differentiation. Basically all the mathematical expressions of deep learning models can be represented as computational graphs, which consist of nodes and edges. Partial derivatives on each edge of a computational graph can then be obtained. With the partial derivatives, we can let software compute differentiation of any node with respect to any variable by utilizing chain rule of Calculus. First of all, the convenience of coding is in the order of CNTK, Tensorflow, and Theano. The criterion is simply based on the lengths of the codes and the learning curve and the ease of coding are not the main concern. According to the criteria, Theano was the most difficult to implement with, and CNTK and Tensorflow were somewhat easier. With Tensorflow, we need to define weight variables and biases explicitly. The reason that CNTK and Tensorflow are easier to implement with is that those frameworks provide us with more abstraction than Theano. We, however, need to mention that low-level coding is not always bad. It gives us flexibility of coding. With the low-level coding such as in Theano, we can implement and test any new deep learning models or any new search methods that we can think of. The assessment of the execution speed of each framework is that there is not meaningful difference. According to the experiment, execution speeds of Theano and Tensorflow are very similar, although the experiment was limited to a CNN model. In the case of CNTK, the experimental environment was not maintained as the same. The code written in CNTK has to be run in PC environment without GPU where codes execute as much as 50 times slower than with GPU. But we concluded that the difference of execution speed was within the range of variation caused by the different hardware setup. In this study, we compared three types of deep learning framework: Theano, Tensorflow, and CNTK. According to Wikipedia, there are 12 available deep learning frameworks. And 15 different attributes differentiate each framework. Some of the important attributes would include interface language (Python, C ++, Java, etc.) and the availability of libraries on various deep learning models such as CNN, RNN, DBN, and etc. And if a user implements a large scale deep learning model, it will also be important to support multiple GPU or multiple servers. Also, if you are learning the deep learning model, it would also be important if there are enough examples and references.

• Radiation Oncology Journal
• /
• v.21 no.3
• /
• pp.183-191
• /
• 2003

Purpose: To suggest the optimal radiation fields after a surgical resection based on a nationwide survey on the principles of radiotherapy for rectal cancer in the Korean Patterns on Care Study. Materials and method: A consensus committee, composed of radiation oncologists from 18 hospitals in Seoul Metropolitan area, developed a survey format to analyze radiation oncologist's treatment principles for rectal cancer after a surgical resection. The survey format included 19 questions on the principles of defining field margins, and was sent to the radiation oncologists in charge of gastrointestinal malignancies in all korean hospitals (48 hospitals). Thirty three ($69\%$) oncologists replied. On the basis of the replies and literature review, the committee developed guidelines for the optimal radiation fields nor rectal cancer Results: The following guidelines were developed: superior border between the lower tip on the L5 vertebral body and upper sacroiliac joint; inferior border $2\~3$ cm distal to the anastomosis in patient whose sphincter was saved, and $2\~3$ cm distal to the perineal scar In patients whose anal sphincter was sacrificed; anterior margin at the posterior lip of the symphysis pubis or $2\~3$ cm anterior to the vertebral body, to include the internal iliac lymph node and posterior margin $1.5\~2$ cm posterior to the anterior surface of the surface, to include the presacral space with enough margin. Comparison with the guidelines, the replies on the superior margin coincided in 23 cases ($70\%$), the inferior margin after sphincter saving surgery in 13 ($39\%$), the inferior margin after adbominoperineal resection in 32 ($97\%$), the lateral margin in 32 ($97\%$), the posterior margins in 32 ($97\%$) and the anterior margin in 16 ($45\%$). Conclusion: These recommendations should be tailored to each patient according to the clinical characteristics such as tumor location, pathological and operative findings, for the optimal treatment. The adequacy of these guidelines should be proved be following the Korean Patterns of Care Study.

• Radiation Oncology Journal
• /
• v.13 no.1
• /
• pp.9-18
• /
• 1995

Purpose : A retrospective analysis was performed to investigate the proper management of maxillary sinus carcinoma. Materials and Methods : Authors analysed 33 patients of squamous cell carcinoma of maxillary sinus treated at Chonnam University Hospital from January 1986 to December 1992. There were 24 men and 9 women with median age of 55 years. According to AJCC TNM system of 1988, a patient of T2, 10 patients of T3 and 22 patients of T4 were availalbe, respectively. Cervical lymph node metastases was observed in 5 patients(N 1;4/33, N2b; 1/33). Patients were classified as 3 groups according to management method. The first group, named as 'FAR' (16 patients), was consisted or preoperative intra-arterial chemotherapy with 5-fluorouracil(5-FU;mean of total dosage;3078mg) through the superficial temporal artery with concurrent radiation(mean dose delivered:3433cGy, daily 180-200cGy) and vitamin A(50,000 IU daily), and followed by total maxillectomy and postoperative radiation therapy(mean dose;2351cGy). The second group, named as 'SR'(7 patients), was consisted of total maxillectomy followed by postoperative radiation therapy(mean dose 5920 cGy). The third group, named as 'R'(6 patients), was treated with radiation alone(mean dose;7164cGy). Kaplan-Meier product limit method was used for survival analysis and Mantel-Cox test was performed for significance of survival difference between two groups. Results : Local recurrence free survival rate in the end of 2 year was $100\%$, $50\%$ and $0\%$ in FAR, SR and R group, respectively. Disease free survival rate in 2 years was $88.9\%$, $28.0\%$ and $0\%$ in FAR, SR and R group, respectively. Overall survival rate in 2 years was $88.9\%$, $40\%$ and $50\%$ in FAR, SR and R group, respectively. There were statistically significant difference between FAR and SR or FAR and R group in their local recurrence free, disease free and overall survial rates. But difference of each survival rate between SR and R group was not significant. Conclusion : In this study FAR group revealed better results than SR or R group. In the future prospective randomized study is in need.

• Horticultural Science & Technology
• /
• v.29 no.5
• /
• pp.456-460
• /
• 2011

This study was carried out to compare the pattern of mutant variation and to evaluate the characteristics of mutants obtained by gamma irradiation in rose 'Kardinal'. Forty four rooted cuttings of 'Kardinal' were irradiated at 70 Gy gamma-ray dose from a $^{60}Co$ source to induce mutants in 2002. The irradiated plants were planted in field, and observed spotting of petal color mutants from 2002 to 2004. Four different kinds of mutant twigs with each different color flower were obtained from the irradiated 'Kardinal' with red petal. After being identified to be a stable mutant from 2004 to 2008, each mutant line propagated by cutting was hydroponic-cultured to evaluate the characteristics in the greenhouse from 2008 to 2009. Four mutant lines obtained from 'Kardinal' with red petal (Red group, 44A, 45B) include KA1 with light pink petal (Red group, 55B-55D), KA2 with pink petal (Red group, 63A-63B), KA3 with deep pink (Red purple, N57A-N57C), and KA4 with orange red (Red group, 43A-43B). Diameters of each flower in four mutant lines were different from 'Kardinal'. The line KA1 was 9.5 cm wide, and it showed the smallest diameter when compared to other mutants. While the line KA2 was the largest one with 12.5 cm 'Kardinal'. Petal number per flower was also variable among the mutants. The line KA2 had 39.8 petals being the largest number among the mutants, while the line KA1 was the lowest one compared to 35.5 petals of 'Kardinal'. Petal color was measured by using colorimeter. Brightness (L) measured at each petal of four mutants increased more than 'Kardinal'. CIE Lab values, a and b decreased more than 'Kardinal' at the petal color of three mutants except the line KA4. Characteristics of shoot, leaf, etc. from four mutants were also different from the ones of 'Kardinal'. The line KA1 was shortest in shoot, node and peduncle length, and lowest in prickle number. The reverse side of leaves was reddish green color in 'Kardinal' as well as the line KA4, but green color in the line KA1, KA2, and KA3.

• Clinical and Experimental Pediatrics
• /
• v.52 no.8
• /
• pp.930-937
• /
• 2009

Purpose : We evaluated the effectiveness of treatment and cardiac complications of replacing a high dose of aspirin with a high dose of ibuprofen for children in acute phase of Kawasaki disease. We also analyzed the possibility of replacing a high dose of aspirin with a high dose of ibuprofen to prevent complications such as Reye󰡑s syndrome caused by aspirin. Methods : One hundred eight children with Kawasaki disease were admitted in the pediatrics department from January 1, 2004 to December 31, 2008. Echocardiography and laboratory tests were performed during diagnosis, and the children were followed-up at 6-8 weeks after the diagnosis. We retrospectively analyzed their characteristics and clinical results. Results : The children were assigned to receive either a high dose of aspirin with intravenous immunoglobulin (IVIG) (aspirin group) or a high dose of ibuprofen with IVIG (ibuprofen group). A total of 55 and 53 children were included in the aspirin and ibuprofen groups, respectively. The mean defervescence period was 6.5${\pm}$2.1 days in the aspirin group, and $6.9{\pm}1.9$ days in the ibuprofen group (P=0.309). The number of failed treatments, during and after treatment, was 8 in the aspirin group and 10 in the ibuprofen group (P=0.547). There were 11 initial cardiac complications in the aspirin group, and 14 in the ibuprofen group, but children who showed improvement at follow-up was 7 and 13 in the aspirin and ibuprofen groups, respectively (P=0.133). Laboratory findings were also improved in both groups. Conclusion : We can be considered whether we will replace high dose of aspirin by high dose of ibuprofen in acute phase of Kawasaki disease. Therefore, we can prevent the severe complications of aspirin use, such as Reye's syndrome.

• The Journal of Korean Society for Radiation Therapy
• /
• v.27 no.2
• /
• pp.107-113
• /
• 2015

Purpose : The measurement of skin dose is very important that treatment of breast cancer. On account of the cold or hot dose as compared with prescription dose, it is necessary to analyse the skin dose occurring during the various plan of the breast cancer treatment. At our hospital, we want to apply various analyses using a diversity of dosimeters to the breast cancer treatment. Subjectss and Methods : In the study, the anthropomorphic phantom is used to find out the dose difference of the skin(draining site), scar and others occurring from the tangential treatment plan of breast cancer. We took computed tomography scan of the anthropomorphic phantom and made plans for the treatment planing using open and wedge, Field-in-Field, Dose fluence. Using these, we made a comparative analysis of the dose date points by using the Eclipse. For the dose comparison, we place the anthropomorphic phantom in the treatment room and compared the measurement results by using the TLD and MOSFET on the dose data points. Results : On the central point of treatment planing basis, the upward and downward skin dose measured by the MOSFET was the highest when the fluence was used. The skin dose of inner and outer was distinguished from the figure(5.7% ~ 10.3%) when the measurements were fulfilled by using TLD and MOSFET. The other side of breast dose was the lowest in the open beam, on the other hand, is highest in the Dose fluence plan. In the different kinds of treatment, the dose deviation of inner and outer was the highest, and so this was the same with the TLD and MOSFET measurement case. The outer deviation was highest in the TLD, and the Inner'was highest in the MOSFET. Conclusion : Skin dose in relation to the treatment plan was the highest in the planing using the fluence technique in general and it was supposed that the high dose had been caused by the movement of the MLC. There's some differences among the all the treatment planning, but the sites such as IM node occurring the lack of dose, scar, drain site are needed pay close attention. Using the treatment planning of dose fluence is good to compensate the lack of dose, but It increases the dose of the selective range rather than the overall dose. Therefore, choosing the radiotherapy technique is desirable in the lights of the age and performance of the patient.

• Korean Journal of Medicinal Crop Science
• /
• v.8 no.3
• /
• pp.234-242
• /
• 2000

There are two cultivars, 'Jeongsun' and 'Mishima' of Bupleurum falcatum L., a medicinal plant, have been cultivated in Korea. Two cultivars were evaluated and compared in details for major growth characters and their seasonal changes. Jeongsun was higher in stem length and root weight per plant, and earlier in maturity than Mishima, while Mishima was higher in top weight per plant and the ratio of outer phloem layer(OPL) to whole root, and more in branch number. Coefficients of variations of most growth characters were higher in Mishima than in Jeongsun. In relationship between root and shoot growth characters, root weight, root diameter, etc. were positively correlated with stem thickness, top weight, branch number, etc. in both cultivars. In comparison of seasonal changes for growth characters between two cultivars, stem length and node number were almost not increased after September, but root length and branch number tended to increase until November. There were continuous increases of top and root weights in both cultivars after September. Increase rates of top and root weights were higher in Mishima and Jeongsun, respectively. The weight ratio of root part to whole plant in Jeongsun and the ratio of OPL to whole root in Mishima were higher during whole growth stages; especially Mishima showed their continuous increases after October. Jeongsun flowered earlier (August 17) than Mishima (September 16) by one month. Complete flowering date was on August 30 in Jeongsun and on September 30 in Mishima. Basic difference in plant growth between two cultivars was maturity, so shoot growth almost stopped after October in Jeongsun, and continued vegetative growth until November in Mishima with late flowering habit. These results indicate that higher root weight in Jeongsun is due to its earlier flowering character leading to relatively longer time for root growth, as compared with Mishima.

• Journal of Gastric Cancer
• /
• v.9 no.4
• /
• pp.207-214
• /
• 2009

Purpose: Subtotal distal gastrectomy has been accepted as the standard treatment for early gastric cancer that's developed on the gastric body. EMR and ESD have been introduced to minimize the incidence of postgastrectomy syndrome, but these procedures can not detect lymph node metastasis and they have a risk for gastric perforation. Segmental gastrectomy has recently been applied for treating early gastric cancer, but its usefulness has not been clarified. The aim of this study was to compare segmental gastrectomy and distal gastrectomy with Billroth I reconstruction for treating early gastric cancer that's developed on the gastric body. Materials and Methods: We performed a retrospective review of all the patients who were diagnosed as having early gastric cancer that developed on the gastric body at Chungnam National University Hospital from January 2004 through July 2007. During this period, 41 patients received segmental gastrectomy and 40 patients underwent subtotal distal gastrectomy. All the patients were studied via a biannual review of the body systems, a physical examination, endoscopy, computed tomography and the laboratory findings. Results: There were no significantly differences of the clinicopathologic characteristics between the two groups. The changes of the nutritional status (Hb, TP, Alb and TC) and the body weight change were not significantly different between the 2 groups. There were significantly more residual food in the SG group than that in the SDG group (RGB classification, Residual>Grade 2), but there were no differences for epigastric discomfort (P>0.05). Esophagitis developed at a similar rate for both two groups (LA classification, >Grade A), and bile reflux was found in only one patient of each group. Conclusion: We expected the reduction of esophagitis and gastritis and the improvement of nutritional status according to the type of procedure. Yet the results of our study showed no significant differences between the two study groups. More patients and a longer follow up time are needed for determining the advantage sand disadvantages of segmental gastrectomy.

• Nuclear Medicine and Molecular Imaging
• /
• v.43 no.1
• /
• pp.26-34
• /
• 2009

Purpose: The purpose of this study was to find out what clinicopathologic or immunohistochemical parameter that may affect FDG uptake of primary tumor in PET/CT scan of the gastric carcinoma patient. Materials and Methods: Eighty-nine patients with stomach cancer who underwent pre-operative FDG PET/CT scans were included. In cases with perceptible FDG uptake in primary tumor, the maximum standardized uptake value (SUVmax) was calculated. The clinicopathologic results such as depth of invasion (T stage), tumor size, lymph node metastasis, tumor differentiation and Lauren's classification and immunohistochemical markers such as Ki-67 index, expression of p53, EGFR, Cathepsin D, c-erb-B2 and COX-2 were reviewed. Results: Nineteen out of 89 gastric carcinomas showed imperceptible FDG uptake on PET/CT images. In cases with perceptible FDG uptake in primary tumor, SUVmax was significantly higher in T2, T3 and T4 tumors than T1 tumors ($5.8{\pm}3.1$ vs. $3.7{\pm}2.1$, p=0.002). SUVmax of large tumors (above or equal to 3 cm) was also significantly higher than SUVmax of small ones (less than 3 cm) ($5.7{\pm}3.2$ vs. $3.7{\pm}2.0$, p=0.002). The intestinal types of gastric carcinomas according to Lauren showed higher FDG uptake compared to the non-intestinal types ($5.4{\pm}2.8$ vs. $3.7{\pm}1.3$, p=0.003). SUVmax between p53 positive group and negative group was significantly different ($6.0{\pm}2.8$ vs. $4.4{\pm}3.0$, p=0.035). No significant difference was found in presence of LN metastasis, tumor differentiation, Ki-67 index, and expression of EGFR, Cathepsin D, c-erb-B2 and COX-2. Conclusion: T stage of gastric carcinoma influenced the detectability of gastric cancer on FDG PET PET/CT scan. When gastric carcinoma was perceptible on PET/CT scan, T stage, size of primary tumor, Lauren's classification and p53 expression were related to degree of FDG uptake in primary tumor.

• Nuclear Medicine and Molecular Imaging
• /
• v.43 no.1
• /
• pp.40-47
• /
• 2009

Purpose: Differentiated thyroid cancer (DTC) has variable degree of F-18 FDG avidity. The purpose of this study was to evaluate the relationship between F-18 FDG uptake and pathological or immunohistochemical features of DTC. Materials and Methods: DTC patients who underwent both pre-operative F-18 FDG PET/CT scan and surgery were included in the study. Maximum standardized uptake values (SUVmax) of primary tumor were calculated. If the primary tumor showed no perceptibly increased F-18 FDG uptake, region of interest was drawn based on finding of a portion of the PET/CT images. Pathological and immunohistochemical markers such as presence of lymph node (LN) metastasis and underlying thyroiditis, tumor size, Ki-67 labeling index, expressions of EGFR, COX-2, and Galectin-3 were evaluated. Results: Total of 106 patients was included (102 papillary carcinomas, 4 follicular carcinomas). The mean SUVmax of the large tumors (above 1 cm) was significantly higher than the mean SUVmax of small (equal to or less than 1 cm) ones ($7.8{\pm}8.5$ vs. $3.6{\pm}3.1$, p=0.004). No significant difference in F-18 FDG uptake was found according to the presence or absence of LN metastasis and underlying thyroiditis, or the degree of Ki-67 labeling index, expression of EGFR, COX- 2 and Galectin-3. Conclusion: In conclusion, the degree of F-18 FDG uptake in DTC was associated with the size of primary tumor. But there seem to be no relationship between F-18 FDG uptake of DTC and expression of Ki-67, EGFR, COX-2 and Galectin-3.